Hi Shaping,
Ha, lots of good questions – nice.
I’ll try to answer both emals inline, so, don’t stop after the first :) ).
Von: Shaping <shap...@uurda.org>
Gesendet: Donnerstag, 4. August 2022 12:01
An: 'Any question about pharo is welcome' <pharo-users@lists.pharo.org>; 'Pharo
Development List' <pharo-...@lists.pharo.org>
Betreff: [Pharo-users] Re: [Pharo-users]Porting from VW 8.3 to Pharo: Pharo100
fileOutValues
Hi Christian. (You can find some other discussion of this in Discord with
Yanni and others.)
I will look, although this kind of always available on demand thing is too
disruptive for me…
I just evaled:
SmalltalkTransform.Pharo100 fileOutValues
This wrote file Values.Pharo100.st of size 42 KB to my VW image directory.
So that little example is a test that shows how the transformation is done. It
converts just package Values to a Pharo-compatible file-in. My task is then to
queue a bunch of ProjectChange instances like this one:
SmalltalkTransform.Pharo100>>
ValuesProject
^ProjectChange
name: #Values
source: (Array with: (Package name: #Values))
changes: (Valuemap with: 'Values' -> self
ValuesTransform)
Exactly
but for my own packages. No bundles are transformed (just their contained
packages) because Pharo doesn’t have bundles.
Is that right?
No, bundles are handled. For real examples, you need to look at the PDFtalk
transforms.
Yes, Pharo does not have a concept of bundles (ordered aggregates of packages).
Instead it relies on a naming convention for packages. That convention is
honored in the fileout, so that packages will be partly grouped in Pharo
according to the category prefix.
For each VW-package, one Pharo package is created. A bundle itself is also
represented as Pharo package with one class About<bundlename> with class
methods for the metadata of the bundle, including a method giving you the
ordered list of component packages. So, all contents and metadata of packages
and bundles are transformed for Pharo. No code or info gets lost.
Is method
ValuesTransform
^PackageChange
ignoredNames:
#(#{Smalltalk.GeneralBindingReference})
bridgeClasses: (Valuemap
with: #{Timestamp} ->
#DateAndTime
with: #{Smalltalk.ColorValue}
-> #Color)
localChanges: self valuesLocalTransform
extensions: (Array
with: (SystemClassChange
className:
#Color
instanceChanges: (Array with: (Add method: #asColorValue code:
#_ph_asColorValue)))
with: (SystemClassChange
className:
#TextStream
instanceChanges: (Array with: (Add method: #nextPutAllText: code:
#_ph_nextPutAllText:))))
written specifically for that package? I would think it applies to all
packages. I see some expected mappings like Timestampe to DateAndTime.
Yes, this method returns a PackageChange Value describing the transformations
needed to create the Pharo fileout for this specific package (inspect the
return value for the fully expanded Value). Methods exist with the same name
for other Smalltalks. Depending on the dialect (or version of a dialect), the
transforms are different. Squeak and Pharo are quite similar, because they
share a common history, but VA or Gemstone need quite different transforms.
So, in general, for each package, there is one such method/Value for each
target Smalltalk/version. I do not dare to extract commonalities before the
machinery is really robust and stable. For now everything is neatly separate
and self-contained (and probably it will stay that way, although there are lots
of duplications).
The mapping of class names is the responsibility of the enclosing ProjectChange
Value where you define the list of source bundles/packages to transform, the
PackageChanges for all packages and the mapping of “global” names.
(The bridge classes above are no renames, but a subclass relationship (is-a) to
avoid renamings. The new class Timestamp will be created as subclass of
DateAndTime which has almost the same semantics. Therefore, I can still use
Timestamp which will be basically a DateAndTime now.)
There is still a technical challenge here. Currently, a ProjectChange need to
include all prerequisites (Values is part of the PDFtalk project and will be
transformed with it). A ProjectWriter, which coordinates the transform, keeps
track of the mappings when they are created (either explicitly or through a
namespace renaming – see implementers of #PDFtalkProject).
I would like to have this more modular: the mappings from the Values
transformation should be persistently saved, so that other transformation
projects can just use them, instead of including the sources into one own
project.
For this, I need to have renamings local to a package (where they first occur),
not global on the project level.
For Values and Values Tests and Values Tools this works, because there are no
mappings in the Values package.
What about conversion of VW arrays to Pharo literal arrays? How is that done?
(I think you mean dynamic arrays like {1. ‘abc’ size. 42} in which evaluation
happens (in contrast to literal things which can be resolved already by the
compiler).
Not! Since a while, VW also has dynamic arrays, but not in VW 8.3 – the last
publicly available version. I will not shut out those users, because
“open-source” would be quite absurd, if it is only available for paying
customers.
In 8.3, the compiler does not accept that syntax and therefore, there is no
easy way to represent this in replacement code.
So, no. It is not possible until Cincom releases an public version which can
handle that.
I recall that one of the Smalltalks (I don’t recall which) had Stream semantics
differing from VW’s.
… I just checked. VW’s #upTo: method includes the object and leaves the index
after it, and Pharo’s excludes the object and leaves the index at the object.
So that is some major breakage if we don’t correct it. Can it be done
automatically?
Yes, these are the usual porting challenges and exactly the reason why this
library exists :). Thank you for the question :).
Yes, the stream semantics need to be fixed. The idea is that a set of
transforms for this issue can be reused by others.
>> valuesLocalTransform
has lots of juicy bits. But this doesn’t look very simple. We can’t just
replace an old method with a new one. We also have to write the new one to
tweak how the indices are used in #upTo:, and make sure that new method gets
filed-in as well into the Pharo target image. Or, we have to do this kind of
change manually.
Naa, it’s very easy, I think :).
A PackageChange specifies transforms for classes used in the package
(#localChanges) and #extensions for system classes of the target. For a class,
you can have a ClassChange describing the changes to instance or class methods.
A MethodChange has 4 subclasses for:
* Ignore – don’t write this method to the target
* Add – add this new method (not in the source system) to the target
* Replace – replace the body of this method with other code
* Rewrite – rewrite the method source using a rewrite rule.
Add and Replace need the target code. This is stored in another method with a
derived name like #_ph_upTo: . The method name is not important, because only
the body of the method is used. But the name should not be used in the source –
it is just a holder for the replacement code. These methods live in the
specific [<Smalltalk> Fileout <Package>] package.
There are lots of working(!) examples for all of those in the PDFtalk transform
project.
How did you handle conversion to literal arrays conversion and Stream semantic
adjustment?
This bit
(SystemClassChange
className:
#Color
instanceChanges: (Array with: (Add method: #asColorValue code:
#_ph_asColorValue)))
is replacing #asColorValue with #_ph_asColorValue because some special
Pharo-color conversion needs to happen. But how does #_ph_asColorValue get
defined? It’s neither in VW nor in Pharo 10.
You got bitten by the old version of [Pharo Fileout Values]. Please load [Pharo
Fileout PDFtalk]. There, the methods exist.
Shaping
[…]
How does one generate the source (run the transformation)? Where are the
aforementioned “defined code transformations”, Is there an example of how to
setup the declarative template needed to map VW namespace names to Pharo class
prefixes?
I loaded the following packages into my VW 8.3 image:
Values Project Yes
Values
Values Fileout Pharo
Values Testing
Values Tools
Values Tools Testing Automatic
Load PDFtalk Project bundle. Since the main target is PDFtalk. Values (and
PostScript) are just bycatch.
Pharo Fileout Values No, this package has been renamed to [Pharo Fileout
PDFtalk]. Load this.
Pharo Transform No, not separate. This is part of the Transform Project.
Smalltalk Transform Project Yes
Smalltalk Transform TestBundle
Smalltalk Transform Tests
Smalltalk Transform
Smalltalk Transform Model
Smalltalk Transform Testing
Smalltalk Transform TestPackage
Smalltalk Transform Tools Automatic
Yes, I just loaded the head packages and the dependencies were also loaded.
Thanks for trimming that down for me.
Any package not updated recently (in the last few months) I didn’t load.
Did I miss anything?
Yes. The ideal setup from a virgin image
Ok. I don’t have a virgin image. I have a very non-virgin image, about 27
years of development I’m trying to port to Pharo. I don’t yet have a specific
interest in the PDFtalk, though I do see a need for PDF generation later, and
will probably revisit that. For now, I just want my own stuff to run in Pharo.
Virgin image just means that you don’t need anything else. You can safely load
it in you favorite special images :).
I would load PDFtalk, although technically you don’t need to (all the
extensions to PDFtalk would be unloadable, but that doesn’t affect Values).
is:
Load {Values Project] bundle
Load {PDFtalk Project} bundle
Load {Smalltalk Transform Project} bundle
Load [Pharo Fileout PDFtalk] package
Save, done
Okay, so do I understand correctly that I need to include the PDFtalk stuff
even if I’m not interested in PDFtalk, because that’s where a lot of the
Smalltalk transformation machinery lives? Or is the PDFtalk just being used as
an example for how to do a massive transformation? Or Both?
No, the transformation machinery is fully independent of PDFtalk. I just tried
it. The dependencies are in the specific [Pharo Fileout PDFtalk] package, since
I have already quite a few replacement methods which are extensions to PDFtalk
classes.
PDFtalk is the focus of the project and therefore all issues are solved first
with this library in mind. Therefore, bundles and namespaces are handled, for
example. When you study the more interesting transformations for PDFtalk, it
would be a shame not to be able to browse the methods and classes involved.
So, PDFtalk is the real world reference example.
To transform Values do: “Pharo100 fileOutValues”
The [Pharo Fileout PDFtalk] package includes the latest Values transformations.
I am thinking about a better modularization…
Also, the wiki is a bit out of control. It really needs some restructuring.
In the cites wiki page, there is a link to a blog where I record the changes.
This might be informative.
2. Port the Values package. This is easy, since no namespaces are involved.
This first instruction after VW package setup says to port the contents of the
Values package from VW to Pharo. Do you mean manually? Probably not.
No, no. This has been finished in March.
For each dialect, I have a GitHub repository where I release important
versions: <https://github.com/PortingPDFtalk/PharoPDFtalk>
https://github.com/PortingPDFtalk/PharoPDFtalk. You can find the working port
as first release “
<https://github.com/PortingPDFtalk/PharoPDFtalk/releases/tag/1.3.0.0> Working
version“. There you can download the ported Values fileout with the exact
description with which versions of what it was created.
This should be a good starting example.
Why do I need any new code installed in Pharo before I begin the
transformation, if I’m transforming code from VW to pharo? I’m not
understanding the basic constraints of the problem, even when the detailed
steps are clear.
No, no. You don’t need anything on the Pharo side. The fileouts on GitHub are
the end products of a transformation for people who don’t use VisualWorks, but
want to use Values in Pharo. Or help with PDFtalk by fixing some issues, so
that I can write the transformations.
I’ve done these steps so far:
1. Went to <https://github.com/PortingPDFtalk/PharoPDFtalk>
https://github.com/PortingPDFtalk/PharoPDFtalk.
Mistake :).
You don’t want to look at the unfinished product of the current version of the
incomplete transformations for PDFtalk.
See https://wiki.pdftalk.de/doku.php?id=stateoftheport#pharo-10-0:
Instead, you want to look at the unfinished product of the transformations of
your projects :).
2. Saved down PDFtalk.Pharo100.st into <my Pharo 10 image
directory>/pharo-local/Smalltalk-Transformation. (I figured that was a good
place to save it. If anyone disagrees, or has a better or more conventional
idea about where files should be saved, please say so. I setup a Pharo Git
repo and played with it briefly for the first time yesterday. I’ve used Pharo
off an on for 16 years, but this is the first time I’m making a serious effort
to manage source, and not throw away what I’m working on.)
3. Filed-in PDFtalk.Pharo100.st. This went on for about 7 minutes or so. I
have a hundreds if not over a thousand classes showing in Epicea. Is there
anyway to get Epicea to give me a count of changes with a time-range filter?
4. Deleted (forgot) yesterday’s, old Main practice-repo from both the image
and the drive, and made a new one. I need to add to Main all the packages I
just filed-in, but I don’t see an efficient way to do that. I would like to
use the Add Package button, but this gives a filtered list of available
packages. I can filter subgroups, and then individually select each of the
checkboxes to the left of each package (there is no Ctrl-A [select all] option
here, which seems to be a strange omission given the potentially large number
of packages involved). I see lots of prefixes for the classes just loaded. I
could easily miss something if I filter/select/add one prefix-group at a time.
Is there an easier way? Over in Epicea I don’t see a way to push the loaded
items listed to a specific repo. The first thought I have is to select all
filed-in code artifacts by datetime span. I did that and saved it as an .ombu
file (I have no idea what that is). I don’t see a way to import that .ombu
file into repo Main’s “Working copy” window. It must be easy, but I don’t see
it. Please suggest the best way.
I’d like to know as well – I am not quite familiar with the source management
concepts in Pharo. It is planned to generate Tonel output in the future, but
for now I feel safer with the traditional fileout where I can have doIts. I am
not sure how Tonel reacts to crippled sources, which are normal during the
development of the transformations.
I feel like I’ve missed the main details of how to start the transformation.
I’d like to do a small fully contained test package with no external
dependencies as a first test, probably my Magnitude or Collection packages.
Suggestions are welcome. Thanks for doing this and sharing it.
If I may suggest, I would look at the transformations on the class side of the
SmalltalkTransform.Pharo class: methods #ValuesProject and #PDFtalkProject (I
know, some would detest, that I start methods sometimes with a capital letter…).
I used to be this way. I’m okay with it now, and actually have selectors that
are completely capitalized Acronyms. Times change.
To get a feel for the machinery, you could debug through “Pharo100
fileOutValues” (for Pharo 10.0).
This will be my next step, but I’m trying to understand the big picture of
dependencies. The file-in was clean. I want to commit that before moving on.
As said, Values don’t use namespaces. To see what is involved there, you could
look at #PDFtalkProject for the name mapping.
But beware! The PDFtalk transformation is not finished yet. Still, you could
get some inspiration and can probably reuse some of the rules.
Great that you are trying to use the transformation project. The more use, the
better it should get.
Please share your questions and insights (and successes :)).
I’ve been looking for a tool like this for many years. Thanks again.
Shaping
Thank you :)
Christian
From: christian.haider <christian.hai...@smalltalked-visuals.com
<mailto:christian.hai...@smalltalked-visuals.com> >
Sent: Wednesday, 22 June, 2022 05:42
To: pharo-users@lists.pharo.org <mailto:pharo-users@lists.pharo.org>
Subject: [Pharo-users] [PDFtalk] second fileOut for Squeak and Pharo
With help from the community some issues were fixed which improved the test
statistics nicely.
Check it out:
https://wiki.pdftalk.de/doku.php?id=portingblog#second_pdftalk_fileout_for_squeak_and_pharo
Thanks to everybody involved!
Happy hacking,
Christian
